Brain (1984), 107, 829-854

CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS

by ELIZABETH K. WARRINGTON and T. SHALLICE (From the National Hospital, Queen Square, London WC1N 3BG)

SUMMARY We report a quantitative investigation of the visual identification and auditory comprehension deficits of 4 patients who had made a partial recovery from herpes simplex encephalitis. Clinical observations had suggested the selective impairment and selective preservation of certain categories of visual stimuli. In all 4 patients a significant discrepancy between their ability to identify inanimate objects and inability to identify living things and foods was demonstrated. In 2 patients it was possible to compare visual and verbal modalities and the same pattern of dissociation was observed in both. For 1 patient, comprehension of abstract words was significantly superior to comprehension of concrete words. Consistency of responses was recorded within a modality in contrast to a much lesser degree of consistency between modalities. We interpret our findings in terms of category specificity in the organization of meaning systems that are also modality specific semantic systems.

INTRODUCTION Neuropsychological and neurological data provide an important source of evidence on the cerebral organization of visual , in identifying not only the component processes in the attainment of a meaningful percept, but also the categories into which meaningful percepts are to be classified. It is now generally accepted that a failure of visual identification that cannot be accounted for by impaired sensory processes (or other confounding factors) can occur as a selective deficit. Not only can visual agnosic deficits at the level of perceptual analysis— apperceptive —be distinguished from those of semantic analysis—associa- tive agnosia—there is also evidence of differentiation at a semantic level. Thus for objects, colours, faces, topography, parts of the body, letters and numbers have all been identified as mutually dissociable deficits and hence it can be inferred that the neural substrates of these categories are also different {see Hecaen and Albert, 1978). There have been pointers to suggest that such classification might be taken a stage further and that a more fine grain categorical organization can be observed within the broad category of visual object agnosia. As yet the evidence is sparse and mainly anecdotal. Nielson (1946) described a patient (C.H.C.) who he claimed had considerably greater difficulty in identifying inanimate objects by sight or touch, and foods except by , than in identifying all living things (including flowers). He also 830 ELIZABETH K. WARRINGTON AND T. SHALLICE observed the converse pattern of deficits in a patient (Flora D) who had greater difficulty in identifying 'animate objects' than inanimate objects. Hecaen and de Ajuriaguerra (1956) reported a single case study of a patient with a left hemisphere lesion in whom they observed a category specific visual agnosia; identification of inanimate objects was more impaired than identification of animals. Konorski (1967) extrapolated from such dissociations between cognitive deficits, to distinguish nine 'gnostic fields', one of which was living things. He also speculated that small manipulable objects form a separate category from nonmanipulable objects. More recently Warrington (1981) provided quantitative evidence of a patient (C.A.V.) who had a significantly greater impairment in the identification of the visual re- presentation of concrete concepts as compared with abstract concepts using a word/ picture matching test (T. Shallice and J. McGill, unpublished) in which the abstract words were represented by pictures of concrete objects (e.g. aptitude—man playing a bassoon) of approximately equivalent familiarity and complexity as the pictures representing concrete concepts (e.g. a donkey). The evidence for category specific deficits is more firmly established for auditory-verbal than visual information. Impaired comprehension of colour names and body part names has long been recorded in the neurological literature. Indeed Goodglass et al. (1966), in the first quantitative investigation of semantic word categories, concluded that the selective impairment or preservation of semantic categories occurred not infrequently in an aphasic population. In a single case study Warrington (1975) demonstrated a significantly greater auditory comprehension impairment for concrete words than abstract words and more recently, quantitative evidence of the selective impairment of the comprehension of inanimate object names has been reported (Warrington and McCarthy, 1983). Category specificity, then, has been documented in both the visual and verbal domains, but the relationship between the two is rarely considered. Typically visual object agnosia is viewed only from the perspective of . Yet the relationship between visual 'semantics' (i.e. knowledge of things), and their verbal equivalent, could well have far reaching theoretical implications. It seems fairly certain that there is not a one to one correspondence between one domain and the other. For instance, it is accepted (and it has occasionally been documented) that verbal comprehension can be intact in patients with a visual object agnosia (Rubens and Benson, 1971; Hecaen et al., 1974). Conversely, patients with word comprehen- sion difficulties—transcortical sensory —are not necessarily agnosic for visual stimuli (Benson, 1979; Warrington and McCarthy, 1983). Such observations have usually been interpreted in terms of a unitary semantic system that has been specifically disconnected from 'lower level' modality specific perceptual systems (Geschwind, 1965). There are, however, a number of agnosic phenomena which are difficult to interpret in terms of a single semantic system disconnected from particular input modalities, and more easily interpreted in terms of modality specific semantic systems. Thus certain semantic memory impairments leave the patient with a stable but impoverished semantic processing ability. Only a CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 831 limited set of items can be identified and for other items, at best, only superordinate information is retained (Warrington, 1975; Coughlan and Warrington, 1981). Such a pattern of impairment can occur together with lack of concordance between modalities; it is then most easily accounted for in terms of the degradation of the stored information within two semantic systems rather than a disconnection. Moreover other observations also suggest the existence of independent modality specific semantic systems. Beauvois et al. (1978) have argued that the observations of patients with optic aphasia and tactile aphasia indicate that separate semantic systems for verbal, visual and tactile information exist {see also Beauvois, 1982). In this paper we report 4 patients recovering from a herpes simplex encephalitis in whom clinical observations suggested the selective preservation of the identification of inanimate objects. Our aim in these investigations was to obtain further evidence for the categorical organization and modality specificity of semantic systems.

CASE HISTORIES Case 1 J.B.R., a 23-year-old electronics undergraduate, was admitted to the National Hospital on January 24,1980. Two days previously he had had a grand mal seizure following which he became progressively drowsy. On examination he was pyrexial, his neck was stiff and Kernigs' sign was positive. He was confused and disorientated. EEG showed repetitive complexes bilaterally and a CT scan showed widespread low attenuation, maximal in both temporal lobes, that was considered consistent with herpes simplex encephalitis (see fig.). He gradually became more active and alert but he remained very confused and for a period of one week, three weeks after the onset of his illness, he was observed to eat and drink indiscriminately, including nonfood items. He continued to improve slowly and by August 1980 there were no neurological signs of note other than the cognitive deficits described below (a detailed neurological account of this patient has been given by Greenwood et al., 1983). Psychological test findings. J.B.R.'s scores on intelligence tests are given in Table 1. He was totally disorientated in time and place and he was densely amnesic for ongoing events and for events prior to his illness. On tests for words and faces he scored at a chance level (Table 1). He was unable to identify contemporary famous faces or names. His perceptual skills were considered

TABLE 1. COGNITIVE TEST RESULTS

J.B.R SBY. K.B. ING.

14.4.80 7.8.80 28.6.80 28.4.81 11.6.82 Verbal IQ 102 101 95 70 70 Performance IQ 78 103 93 76 58 Arithmetic 12 15 9 4 4 Similarities 11 10 7 3 6 Digit Span 9 7 7 7 6 Vocabulary 9 8 12 3 4 Picture Completion 7 8 7 6 2 Block Design 9 11 10 5 6 Picture Arrangement 4 7 5 6 0

Matrices IQ 99 100 NT 90 70

Recognition Memory Words (n = 50) 29 - 29 NT 23 Faces (n - 50) 30 18 27 25 Performance on WAIS, Matrices and Warrington's (1984) Recognition Memory Tests. 832 ELIZABETH K. WARRINGTON AND T. SHALLICE to be intact (e.g. on the Warrington and Taylor (1978) test of matching usual and unusual view photographs he scored 20/20 correct and 19/20 correct on the Warrington and James (1967) frag- mented letter test). His spontaneous speech was fluent although he used a somewhat limited and repetitive vocabulary; occasional word-finding difficulty was noted and he tended to use circum- locutory expressions. Articulation, phrase length and syntax were considered to be normal. His score on the modified Token Test (Coughlan and Warrington, 1978) was satisfactory (15/15). He failed to score on a graded difficulty naming test (GNT) (McKenna and Warrington, 1983) and on a test of naming common objects he scored 11/15 (Coughlan and Warrington, 1978). He failed to score on the NART (Nelson, 1982) and on the Schonell reading test he made dyslexic errors (regularizations) with the lower frequency irregularly spelt words. He had no difficulty in naming or identifying colours, shapes and letters. However, his ability to identify inanimate objects by sight or by touch was impaired and he appeared to have more difficulty in identifying pictures of animals than pictures of inanimate objects and in comprehending animal names than object names. He performed at a very defective level on the Peabody Picture Vocabulary Test (score 26). Similarly on the Shallice and McGill Concrete/ Abstract Picture Word Matching Test his comprehension of both categories of word was severely impaired (see Table 2).

TABLE2. CONCRETE/ABSTRACT WORD/PICTURE MATCHING Concrete Abstract n = 30 n = 30 J.B.R. (7.4.80) 14 19 S.B.Y. (28.6.82) 15 11 L. hemisphere lesion group 1.6 7.6 Error scores for J.B.R. and S.B.Y. and the mean error score for a consecutive series of patients with left hemi- sphere lesions (n = 45) (chance = 22.5).

In summary, J.B.R. was able to function at an overall average level on tests of intelligence. He was densely amnesic. His language skills, especially his spontaneous speech, were relatively intact. He had significant difficulties, however, both with visual identification and word comprehension.

Case 2 S.B.Y., a 48-year-old naval officer (engineer), was admitted to a private hospital in Hong Kong on February 15, 1982. He was febrile and confused. He became comatose and was treated with cytosine arabinoside and dexamethasone. A CT scan was normal but there was a raised protein content in the CSF. His level of gradually returned to normal and on May 7,1982 he was transferred to the Queen Elizabeth Military Hospital, Woolwich. On examination there were no neurological signs other than the cognitive deficits described below. A repeat CT scan in May 1982 showed areas of low attenuation in both temporal lobes that were considered to be compatible with the diagnosis herpes simplex encephalitis (see fig.). Psychological test findings. S.B.Y.'s intelligence test results are given in Table 1. He was almost totally disorientated in time and place and his memory for ongoing events was very impaired. He scored at a chance level on recognition memory tests for words and faces. His perceptual skills appeared to be intact insofar as he made no errors on the fragmented letters test and on matching usual and unusual view objects. His spontaneous speech was very fluent and apart from a tendency to repeat himself, only occasional word-finding difficulties were noted. His phrase length, syntax and articulation appeared to be entirely normal. His performance on the modified Token Test was very satisfactory (15/15). However, his comprehension of individual words appeared to be impaired and on tests of object CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 833 naming his performance was very poor (GNT 0/30) as was naming from description (1/15) (Coughlan and Warrington, 1978). He was still able to identify and name colours and letters and he was able to read at a high average level (NART 31/50). Identification of inanimate objects by sight and by touch was impaired and he appeared to have more difficulty with animal and food names than object names. His performance on the Peabody Picture Vocabulary Test was very weak (score 73). On the Shallice and McGill Concrete/Abstract Picture Matching Task, although his score was only slightly weak compared with patient controls for the abstract words, it was very significantly impaired for the concrete words (Table 2). In summary S.B.Y.'s overall level on intelligence tests was just within the average range. It is assumed that he would have been able to function at least within the high average range premorbidly and that these scores indicate some generalized intellectual deterioration. In addition he had a dense amnesic syndrome, a visual identification deficit and impaired word comprehension.

Case 3 K.B., a 60-year-old housewife, was admitted to the National Hospital on April 27, 1981 for further assessment of her cognitive deficits following a herpes simplex encephalitis diagnosed in June 1979. During the acute illness she developed a frontal headache, fever and increasing confusion over a 48 h period. During the early stages of recovery she was reported to be very amnesic, dysphasic and agnosic. She was also observed to eat food and nonfood items indiscriminately. On examination there were no neurological signs of note other than cognitive dysfunctions (described below). An EEG was normal. A CT scan showed marked residual low attenuation bitemporally most marked on the left (see fig.). (A detailed neurological account of this patient has been given by Greenwood et aJ., 1983.) Psychological test findings. Intelligence test results are given in Table 1. She was completely disorientated for time and place and appeared to be almost totally unable to recall or recognize ongoing events. On a recognition memory test for faces she scored at a chance level. Her spontaneous speech was very sparse as to word choice and word-finding difficulties were very evident. On the Boston Diagnostic Aphasia Battery the pattern of deficit was that of a global dysphasic with articulation spared. She was only able to name 1/15 common objects and she failed to score on a test of naming from description. Her performance on the Peabody Picture Vocabulary Test was extremely poor (score 26). It was noted that she was unable to name or identify objects by touch. She was fairly accurate in identifying and naming letters, actions, body part names and colours. Her reading and spelling were impaired (she failed to score on the NART). In the course of clinical assessment it was observed that K..B. had more difficulty in identifying animals and foods than inanimate objects and a similar pattern of difficulty was observed for her word comprehension.

Case 4 I.N.G., a right-handed housewife aged 44 years, was admitted to the National Hospital between June 10 and 27, 1982 for assessment of her suitability for rehabilitation. A herpes simplex encephalitis had been diagnosed in November 1978 on the basis of a three day history of vomiting, pyrexia and seizures. A CT scan then showed a low attenuation lesion deep in the right hemisphere and a brain biopsy demonstrated necrotic tissue. At the time of her discharge in June 1979 she was severely amnesic, aphasic and agnosic and she had been observed to attempt to eat nonfood items. She was still incontinent and continued to have generalized seizures which were preceded by left-sided clumsiness and turning of her head to the right. On admission to the National Hospital, neurological examination was normal apart from higheT cerebral function deficits. A repeat CT scan showed areas of bilateral temporal low attenuation and marked atrophy which were considered consistent with her known herpes encephalitis {see fig.). Psychological test findings. Intelligence test results are given in Table 1. She was disorientated in time and place and was completely unable to recall or recognize ongoing events or hospital personnel. On recognition memory tests for words and faces she scored at a chance level (Table 1). Her spontaneous 834 ELIZABETH K. WARRINGTON AND T. SHALLICE

FIG. CT scans of A, J.B.R. (October 1980). B, S.B.Y. (May 1982). c, B.A.R. (April 1981) and D, I.N.G. (June 1982). speech was exceptionally sparse; she tended to repeat automatic phrases and very rarely attempted propositional speech. Within these limitations her phrase length, syntax and articulation appeared to be normal. Her performance on object naming tests was very impaired (e.g. 5/15 common objects correct) as was her performance on the modified Token Test (9/15). However, she could name the primary colours and common shapes. Her reading and spelling skills were extremely defective. Her ability to identify common inanimate objects by sight or touch was impaired and she appeared to have even greater difficulty in identifying foods and animals. She performed at a very defective level on the Peabody Picture Vocabulary Test (score 44). I.N.G. was very distractable in the test situation and it was somewhat difficult to hold her attention for sufficient time to complete tests without interruption. In summary, there was evidence of a fairly marked degree of generalized intellectual impairment In addition she was markedly amnesic, dysphasic, dyslexic, dysgraphic and agnosia

Summary of Case Reports These 4 patients were at various stages of recovery from a herpes simplex encephalitis. The clinical course of their illness, the neurological findings and the residual cognitive deficits were qualitatively CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 835

similar in the 4 cases. CT scans demonstrated predominantly bitemporal damage in all 4 patients. In addition to varying degrees of dementia and expressive speech functions, these patients all had a severe global amnesic syndrome and a comprehension deficit which encompassed both visual and verbal knowledge. In all 4 cases clinical observation suggested that within the domain of 'concrete' concepts, knowledge of inanimate objects was spared relative to knowledge of living things and food. Their residual comprehension capacities were documented in more detail.

EXPERIMENTAL INVESTIGATIONS Difficulties in visual identification had been observed in all 4 patients on clinical assessment and in addition difficulties in verbal comprehension were apparent. Our aim in this series of experiments was to document and explore these impairments and in particular, to consider the possible categorical specificity of their deficits and the correspondence between the visual and verbal deficit. The main clinical difference between these 4 patients was in their expressive speech capacities. Neither K.B. nor I.N.G. were able to give coherent word definitions or verbal descriptions of pictures and they were therefore tested in one experiment only (Experiment 8) in 836 ELIZABETH K. WARRINGTON AND T. SHALLICE which a matching to sample technique was adopted. J.B.R. and S.B.Y. were both able to express themselves fairly lucidly and fluently within the constraints imposed by their impoverished verbal vocabulary. It was therefore possible to investigate their comprehension capacities in much more detail using picture description and word definition tasks (Experiments 1-7). In these experiments each verbal response was scored correct or incorrect by three independent judges and the analyses were based on the consensus rating. A lenient scoring criterion was adopted, the judges being instructed to accept as correct 'reasonable evidence that the core concept is conveyed'. In addition they were required (1) to make allowance for dysphasic word-finding difficulties, (2) to credit any correct homonym, (3) to ignore the evidence of gesture, (4) not to give credit for superordinate information alone, (5) to ignore verbal description of features present in a picture (e.g. tail, legs) and (6), in the case of picture description, identification was assumed to be correct if the was correctly named.

Experiment 1: Identification of Visual and Verbal Stimuli Our aim in this experiment was to document the relative severity and concordance of J.B.R. and S.B.Y.'s impaired identification of visual and verbal stimuli. The test stimuli consisted of 40 clear line drawings (mostly of inanimate objects) selected from the Zinkin and Birtchnell set (1968). All items were well within the visual/ verbal vocabulary of the average adult. The patients first attempted to identify the pictures by naming or description. After each response, irrespective of quality, the patient was asked to elaborate once. After a short interval the patients attempted to define the object names presented auditorially. S.B.Y. attempted this task a second time ten days later and J.B.R. was retested on the visual version only, a second time, later on the same day. The percentage of responses judged correct for the visual and verbal version of this task for each patient, together with the number of visual stimuli correctly named, is given in Table 3. There was a small but insignificant advantage of auditory presentation for both patients (McNemar test P > 0.2 for all comparisons). In this and succeeding experiments stability in performance over repeated testing was assessed using the contingency coefficient (Siegel, 1956) with the proviso that

TABLE 3 EXPERIMENT 1. IDENTIFICATION OF VISUAL AND VERBAL STIMULI Visual Auditory J.B.R. (20.6.80) 68 78 (20.6.80) 68 NT S.B.Y. (28.6.82) 50 53 (6.7.82) 48 55 Percentage correct for each presentation condition on each occasion. CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 837 performance for both sets of data lay between 20 and 80 per cent (this avoids artefacts arising from floor and ceiling effects). Comparing responses within a modality, for J.B.R. there was a significant degree of consistency between the two sets of responses in the visual modality (C = 0.61, P < 0.001) and S.B.Y. showed the same consistency effect in both modalities (visual, C = 0.48, P< 0.001; verbal, C = 0.33, P < 0.05). It should be noted that the maximum value for a 2x2 contingency coefficient (C) is 0.71. By contrast, the pattern of consistency scores between modalities was on the whole weaker. For J.B.R. there was an insignificant degree of consistency between his ability to identify an item presented visually and the same item presented auditorially (C = 0.26, P > 0.05). This was also the case for S.B.Y. on the first test session (C = 0.15, P > 0.2); on the second occasion, however, there was a significant degree of concordance (C = 0.49, P > 0.01). Performance on this simple task would be expected to be at ceiling level in any normal average adult. Given the patients relatively intact expressive speech, the high error rate observed on both the visual and verbal version of this task pro- vides quantitative evidence of a moderately severe visual and verbal agnosia in both patients (very comparable in severity to the patients reported by Warrington, 1975).

Experiment 2: Identification of Inanimate Objects and Living Things It had been observed that the ability of these patients to identify certain categories of both picture stimuli and 'concrete' words, namely, animals and plants, was particularly poor whereas inanimate objects appeared to be relatively spared. The aim of this experiment was to explore the possibility of a category specific identification deficit for both visual and verbal stimuli. The test stimuli consisted of 96 coloured pictures: 48 animals and plants and 48 inanimate objects. The two sets did not differ significantly in spoken word frequency (see Appendix). J.B.R. and S.B.Y. first attempted to identify by naming or describing each stimulus picture, alternating between the two major categories. Secondly, they attempted to define each picture name, presented in the same order, auditorially. The percentage correct for each category for each mode of presentation for each patient is given in Table 4. For both J.B.R. and S.B.Y. there was a striking discrepancy between their ability to identify inanimate objects either with pictures or words and their inability to identify living things. J.B.R. was almost at ceiling level on the visual inanimate object condition, yet he virtually failed to score on the visual living things condition. It was also shown that his ability to name objects that he succeeded in identifying was relatively well preserved and far superior to his ability to name living things; but this was not the case for S.B.Y. whose naming skills were very impaired (see Table 4). Although S.B.Y.'s overall level was somewhat worse than J.B.R.'s both patients were usually able to identify the superordinate category with auditory presentation (see Table 4). The small difference in favour of picture identification compared with name definition for 838 ELIZABETH K. WARRINGTON AND T. SHALLICE

TABLE 4. EXPERIMENT 2. IDENTIFICATION OF PICTURES, NAMES, LIVING THINGS AND INANIMATE OBJECTS

Living things Inanimate objects

Visual Auditory Visual Auditory Identified Named Identified Superordinate Identified Named Identified Superordinate J.B.R. (5.8.80) 6 6 8 90 90 67 79 94 O n y (27.7.82) 0 0 0 75 75 0 52 85 Percentage correct identification score, naming score and superordinate score. inanimate objects was significant for S.B.Y. (McNemar test A =16, D = 5, P < 0.05) but not for J.B.R. (McNemar test A = 9, D = 4, P > 0.1). Qualitatively the difference between the major categories was very striking. Some examples of their responses to the inanimate object words were as follows. J.B.R. Tent—temporary outhouse, living home. Briefcase—small case used by students to carry papers. Compass—tools for telling direction you are going. Torch—hand-held light. Dustbin—bin for putting rubbish in. S.B.Y. Wheelbarrow—object used by people to take material about. Towel—material used to dry people. Pram—used to carry people, with wheels and thing to sit on. Submarine—ship that goes underneath sea. Umbrella—object used to protect you from water that comes. In contrast some examples of their responses to living things were as follows. J.B.R. Parrot—don't know. Daffodil—plant Snail—an insect animal. Eel—not well. Ostrich—unusual. S.B.Y. Duck—an animal. Wasp—bird that flies. Crocus—rubbish material. Holly—what you drink. Spider—person looking for things, he was a spider for a nation or country. As there were both floor and ceiling effects it was only possible to compute the concordance of visual and auditory presentation for S.B.Y.'s responses to the inanimate objects category. The contingency coefficient was not significant (C = 0.12, P> 0.2). Experiment 3: Definition of Names of Living Things and Inanimate Objects In the previous experiments we documented a particularly striking dissociation of the ability to identify pictures and names of inanimate objects compared with CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 839 pictures and names of living things. Our aim in this experiment was to assess the reliability of these observations by using a completely different set of words. The test items consisted of names of 40 living things (animals and plants) and 40 inanimate objects matched for frequency. J.B.R. was required to define each name, alternating between the two major categories. This task was attempted twice on the same day (June 20, 1980). S.B.Y. did not attempt this task.

TABLE 5. EXPERIMENT 3. IDENTIFICATION OF NAMES OF LIVING THINGS AND INANIMATE OBJECTS Living things Inanimate objects

1st attempt 2nd attempt 1st attempt 2nd attempt

Identified Superordinate Identified Superordinate Identified Superordinate Identified Superordinale 15 50 10 63 85 93 75 93 Percentage correct identification score and superordinate score for J.B.R.

The percentage correct score for each category for each mode of presentation is given in Table 5. Again there is a very striking discrepancy in the quality of his responses to stimuli in the two categories. Of the living things, he identified only a very few of the high frequency items and was able to indicate the superordinate category for approximately half the stimuli. Identification for most of the inanimate objects was remarkably good and he almost always knew the superordinate category (see Table 5). His ability to indicate the superordinate for the inanimate object category was significantly better than for living things ((i)x2 = 17.64, P < 0.01; (ii) x2 = 10.32, P < 0.01). Our findings in this experiment replicate the verbal condition of Experiment 2. Consistency measures could not be computed because of the floor and ceiling effects.

Experiment 4: Identification of Foods and Inanimate Objects In addition to the selective impairment in the identification of living things, it had been observed that J.B.R.'s comprehension of food names appeared to be very poor. The aim of this experiment was to compare his ability to identify pictures and names of foods and inanimate objects. S.B.Y. was not tested on this experiment. The test stimuli consisted of 30 coloured pictures of foods and 30 coloured pictures of inanimate objects (see Appendix). J.B.R. first attempted to identify each picture by naming or description, the foods and the inanimate object items being alternated. He then attempted to identify each food and object name. The percen- tage correct for each category for each presentation mode together with his naming and superordinate identification score is given in Table 6. Some days later he attempted to mime an action appropriate to 20 foods and 20 inanimate objects (e.g. banana, spaghetti—actual object presented) that partially overlapped with those items previously used. His miming performance was scored jointly by T.S. and E.K.W. 840 ELIZABETH K. WARRINGTON AND T. SHALLICE

TABLE 6. EXPERIMENT 4. IDENTIFICATION OF FOODS AND INANIMATE OBJECTS

Foods Inanimate objects

Visual Auditory Visual Auditory

Identified Named Identified Superordinate Identified Named Identified Superordinate 20 20 30 93 87 40 77 87 Percentage correct identification score, naming score and superordinate score for J.B.R.

His score was 20 per cent correct for foods and 65 per cent correct for inanimate objects. His ability to identify inanimate objects for both modes of presentation is significantly better than his ability to identify foods (visual \2 = 26.8, P < 0.001, 1 d.f.; verbal x2 — 13.1, P < 0.001, 1 d.f.) and an identical pattern of scores was obtained on the mime condition (x2 = 8.29, P < 0.01, 1 d.f.). The miming result provides further evidence that this category specific impairment cannot be attributed to any difficulty of expression, eliminating the possibility that food stimuli are more difficult to describe than inanimate object stimuli. In this experiment we have obtained as striking a dissociation between identification of food and inanimate objects as was obtained for living things and inanimate objects in the previous two experiments. Consistency measures could not be computed because of floor and ceiling effects.

Experiment 5: Identification of Animate, Inanimate Objects and Foods A problem in comparing categories in the visual mode is that items may differ in familiarity just as words do in their frequency of usage. Although word frequency counts are available for all test stimuli used in these experiments, this may not be an adequate control for visual familiarity. The Snodgrass and Vanderwart (1980) stimuli were specifically standardized to provide visual familiarity ratings and were thus considered particularly suitable for investigating both category and mode of presentation effects. The test stimuli consisted of a subset of the Snodgrass and Vanderwart drawings: 40 animals, 40 inanimate objects and 20 foods. S.B.Y. (July 27,1982) first attempted to name or identify each of the 100 drawings, presented in pseudorandom order. Secondly, he attempted to identify each stimulus name (these stimuli were not available to us at the time we tested J.B.R.). The percentage correct scores for each category for each mode of presentation are given in Table 7.

TABLE 7. EXPERIMENT 5. SNODGRASS AND VANDERWART PICTURES TEST Animals Foods Inanimate objects

Visual Auditory Visual Auditory Visual Auditory 13 10 25 25 60 65 Percentage correct for each category for each mode of presentation for S.B.Y. CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 841

For both types of stimuli an analysis of covariance using the linear logistic model was performed. For the visual results there was a highly significant effect of category (x2 = 17.5, P < 0.001, 2 d.f.) with inanimate objects being identified significantly better than both animals (z = 3.3, P < 0.001) and food (z = 2.65, P < 0.01); there was no significant difference between animal and food categories (z = 0.41, P > 0.3). For the auditory modality the results were very similar; there was an overall effect of category (x2 = 23.2, P < 0.001, 2 d.f.) with inanimate objects being identified better than both animals (z = 3.64, P<0.01) and foods (z = 3.13, P< 0.001) with no difference between comprehension of animal and food names (z = 0.41, P > 0.3). The consistency of his visual and verbal responses for the inanimate objects category for the two presentation modes was computed: the contingency coefficient was insignificant (C = 0.04, P > 0.5). In this experiment we have taken into account not only word frequency but also visual familiarity and clear evidence of significant category effects is still obtained. The lack of concordance between the two modes of presentation is again a feature of the data. Experiment 6: Abstract and Concrete Word Comprehension In the previous experiments three specific categories were explored and evidence for category specific deficits within the 'concrete' domain were documented for both a visual and auditory mode of presentation. Identification of living things and foods appeared to be selectively impaired. Our aim in this experiment was to explore the ability of J.B.R. and S.B.Y. to comprehend a wider range of verbal stimuli that spanned the full abstract/concrete range. The test stimuli consisted of two subsets of the Brown and Ure (1969) word list. J.B.R. attempted to define 323 words of which 132 were of AA and A frequency, the remaining 191 were in the frequency range 1 to 50. Half of the total pool had a concreteness rating above the median value of 4.47. S.B.Y. attempted to define 190 words of which 50 were A or A A frequency and 140 in the 1 to 50 frequency range; half of each frequency range had a concreteness rating above the median. In order to ensure correct perception of each word the patients were asked to repeat it correctly before attempting a definition. The percentage correct for each level of concreteness and frequency for each patient is given in Table 8. For both patients there was a significant effect of

TABLE 8. EXPERIMENT 6. ABSTRACT/CONCRETE WORD COMPREHENSION

Concrete Abstract Frequency >4.47 <4.47 J.B.R. (4.8.80) A and AA 89 93 (28.6.82) <50 65 71 S.B.Y. A and AA 76 91 (6.7.82) <50 50 94 Percentage correct for each level of frequency and concreteness. 842 ELIZABETH K. WARRINGTON AND T. SHALLICE

2 2 frequency (J.B.R., x = 20.79, P < 0.001, 1 d.f. and S.B.Y., x = 5.90, P < 0.02, 1 d.f., respectively). There was a highly significant effect of concreteness for S.B.Y. (x2 = 35.3, P< 0.001, 1 d.f.) with performance on abstract words being much superior to that on concrete. This was not the case for J.B.R. (x2 = 1.57, P > 0.2). S.B.Y.'s ability to define abstract words was particularly impressive; some examples of his responses are as follows. Debate Discussion between people, open discussions between groups. Malice To show bad will against somebody. Deceive To let people down—give them the wrong ideas and wrong impression. Caution To be careful how you do something. By contrast some examples of his responses to concrete words are as follows. Ink Food—you put on top of food you are eating—a liquid. Frog An animal—not trained. Cabbage Use for eating, material it's usually made from an animal. Tobacco One of your foods you eat. The patient's incorrect responses were rated by three judges according to the same classification adopted by Coughlan and Warrington (1981) as follows. Omission Made no attempt at definition. Vague Imprecise, residual comprehension indicated (e.g. puppy—'young animal'). Phonological Responses in which the subject appears to be denning a word phonologically similar to the target word (e.g. sap—'kitchen tool device for turning the water on'). Miscellaneous (e.g. mutton 'some sort of rubbish').

TABLE 9. EXPERIMENT 6. ERROR ANALYSIS

Omission Vague Phonological Miscellaneous J.B. R. 4 22 27 47 SB .Y. 0 60 11 30 Percentage correct for each type of error.

The percentage for each type of error is given in Table 9. The most interesting aspect of the error analysis is the occurrence of phonological errors, a phenomenon previously reported by Coughlan and Warrington (1981). As in the case of the previous study these errors cannot be attributed to impairment at an early stage of auditory perception as the subject was required to repeat each target word. The main findingo f this experiment is, however, the evidence for selective impairment of the knowledge of concrete concepts, which has been recorded twice previously (Warrington, 1975, 1981). S.B.Y. showed this dissociation particularly clearly. It should also be noted that he exhibited this same dissociation in picture-word matching (see Case Report). CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 843

Experiment 7: Comparison of Multiple Verbal Categories With verbal stimuli it is feasible to investigate a wide range of categories and so obtain a selection which has not been predetermined by clinical observation. Battig and Montague (1969) have obtained frequency norms for 56 verbal categories; 26 of these were considered appropriate for more detailed investigation of possible category specificity. For the most part exclusion was determined by cultural factors (e.g. US colleges), the lack of sufficient high frequency items (e.g. units of time) or their being proper names (e.g. girls' names). The test stimuli consisted of 12 items of fairly high category-frequency from each of the 26 categories. J.B.R., tested on two days, attempted to define one item from each category in turn and this procedure was repeated until he had attempted all 312 words. (It was not possible to test S.B.Y. on this experiment.) Since this word pool included categories with items that might be difficult for normal subjects to define precisely (e.g. metals, fish), a control subject matched for age and occupation with J.B.R. was also tested. The number correct for each category for J.B.R. and the control subject is given in Table 10. In order to control for word frequency effects an expected score for each category

TABLE 10 EXPERIMENT 7. BATTIG AND MONTAGUE CATEGORY TEST Obtained Expected Difference Control score score score score 1. Clothing 10 8.6 + 1.4 12 2. Dwelling 10 9.5 + 0.5 11 3. Furniture 10 8.5 + 1.5 11 4. Kitchen utensil 10 8.6 + 1.4 12 5. Occupation 10 8.6 + 1.4 11 6. Body part 10 10.3 -0.3 11 7. Vehicle 10 9.4 + 0.6 9 8. Sport 9 8.1 + 0.9 9 9. Weather 9 8.8 + 0.2 9 10. Crime 8 4.7 + 3.3 9 11. Earth formation 8 9.6 -1.6 12 12. Tools 8 8.5 -0.5 11 13. Animals 6 9.5 -3.5 11 14. Science 5 7.1 -2.1 11 15. Insect 3 8.5 -5.5* 10 16. Metal 4 9.6 -5.6* 6 17. Drink 2 6.1 -4.1* 11 18. Cloth 2 8.5 -6.5* 8 19. Musical instrument 2 8.1 -6.1* 12 20. Disease 1 6.7 -5.7* 9 21. Fish 1 7.7 -6.7* 9 22. Precious stone 1 6.1 -5.1* 8 23. Flower 0 5.9 -5.9* 7 24. Fruit 0 7.7 -7.7* 9 25. Tree 0 7.5 -7.5* 8 26. Vegetable 0 6.9 -6.9* 11 J.B.R.'s (5.8.80-8.8.80) and control subject's performance for each category. * Significant difference at P < 0.05 between J.B.R.'s obtained and expected scores. 844 ELIZABETH K. WARRINGTON AND T. SHALLICE was estimated from J.B.R.'s level of performance in Experiment 6 on five frequency bands of the Brown and Ure pool which span the full concrete/abstract range (AA, A, 10 to 49, 1 to 9, < 1). Each individual word was given an 'expected' score depending upon its frequency band; for instance a word in the frequency band 1 to 9 received a score of 0.58, the probability correct for that band in the previous experiment. The sum of these scores was taken as the expected value for that category {see Table 10). For each category the significance of the obtained score and the expected score was computed using an x2 test. For the most part J.B.R. either scored at a level commensurate with his overall level on the mixed abstract/concrete pool or at a very low level indeed. Thus on his ten best categories his score averaged more than 80 per cent correct whereas for his ten worst his score averaged less than 10 per cent correct, far worse than the control subject. These categories do include some that the control subject also found somewhat difficult (e.g. metal, tree). It appeared that the control subject's poor performance on these categories could be attributed to characteristics of this word pool that result in their being somewhat difficult to define other than in terms of the superordinate (e.g. fish) or containing a considerable number of low frequency items (e.g. precious stone). Even allowing for these variations in category difficulty, J.B.R.'s pattern of performance would appear to offer strong evidence for the selective preservation and selective impairment of particular verbal categories. On this account, the distribution of difference scores would be expected to be bimodal, which it appears to be by inspection. To assess this possibility more formally, Dr I. Nimmo-Smith applied Silverman's (1981) simulation procedure for estimating the number of modes in a distribution. The hypothesis that the obtained scores came from a unimodal distribution could be rejected at P < 0.005 significance level. Turning to the individual categories, the findings of this experiment are in general in accord with the category effects reported in Experiments 2, 3 and 4. There are a number of somewhat anomalous findings that will be discussed below. Experiment 8: Recognition of Animals, Foods and Inanimate Objects Clinical assessment of K.B. and I.N.G. indicate fairly marked nominal and expressive difficulties. It was therefore inappropriate to investigate picture and word identification by verbal description. Instead a matching to sample technique was adopted. J.B.R. was also tested. Identification of animals, foods and inanimate objects was tested using a spoken word/picture matching test. The test stimuli consisted of arrays of 5 coloured pictures which were 5 animals, 5 foods or 5 inanimate objects (all taken from the Ladybird books series for young children) and the task was to match the spoken word to one of the 5 items in the array. The patients were tested in blocks of 20 trials; within a block the array remained the same and each of the 5 test stimuli was presented four times in pseudorandom order. The patients were under no pressure to respond quickly; the intertrial interval was approximately 2 s and the interblock interval approximately 2 min. The three stimulus categories were tested in a 3 x 3 Latin Square design, a different set of 5 CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 845 items being used for each array, giving in all 15 items in each category {see Appendix). The percentage correct for each category for each patient is shown in Table 11. For each patient 3x2 contingency tables were drawn up comparing across categories the number of items for which performance was perfect across all four trials with the number where performance was impaired. The contingency tables were then analysed using the Lancaster/Irwin method for partitioning the x2 statistic (Everitt, 1977). For all 3 patients there was no significant difference between performance with animals and foods (J.B.R., x2 = 1.21, P > 0.2, 1 d.f.; K.B. and I.N.G., x2 = 0). For all 3 patients performance on inanimate objects was signifi- 2 cantly better than the other two categories (J.B.R., x = 13.01, P < 0.001, 1 d.f.; 2 2 K.B., x = 18.41, P < 0.001, 1 d.f.; I.N.G., x = 5.41, P < 0.05, 1 d.f.). Matching to sample therefore gives completely analogous results to the verbal description procedure.

TABLE 11. EXPERIMENT 8. WORD/PICTURE MATCHING

Animals Foods Inanimate objects J.B.R. (25.2.83) 67 60 98 KB. (7.5.81) 45 55 85 I.N.G. (14.6.82) 80 85 97 Percentage correct for each category (Chance = 20 per cent).

DISCUSSION In this series of experiments we have attempted to document the visual identification and auditory comprehension deficits of 4 patients who had made a partial recovery from herpes simplex encephalitis. These patients were selected for more detailed investigation on the basis of clinical observations that suggested that their knowledge of certain categories of visual objects appeared to be selectively impaired. It had been noted that they all had much greater difficulty in identifying living things than inanimate objects. The patients differed in one major respect. Two of the patients, J.B.R. and S.B.Y., had fluent spontaneous speech with normal articulation, phrase length and syntax. The other 2 patients, K.B. and I.N.G., however, had marked expressive speech difficulties; only a single experiment using the matching to sample technique was therefore possible. In other respects the pattern of performance of all 4 patients was remarkably similar. Two minor exceptions were noted: (1) the selective preservation of abstract word comprehension in S.B.Y., and (2) the relative preservation of nominal skills in J.B.R. In this discussion we will make the assumption that had it been feasible to test K.B. and I.N.G. using verbal description procedures, their 846 ELIZABETH K. WARRINGTON AND T. SHALLICE performance would in all critical respects have been similar to that of J.B.R. and S.B.Y. The major findings of this investigation are as follows. 1. Category Specificity The identification of inanimate objects was remarkably well preserved when compared with the identification of living things and foods. Experiments 2 and 3 established a very striking dissociation between identification of objects and living things. In Experiment 4 it was found that the identification of foods was very impaired as compared with inanimate objects. In Experiments 5 and 8 the same dissociation between inanimate objects and foods and living things was observed. Furthermore our evidence for category specificity is derived from four different methods of assessment, namely, verbal description, naming, mimed responses and picture/word matching. Other aspects of category specificity were explored in Experiments 6 and 7. Comprehension of abstract words was remarkably well preserved in one patient (S.B.Y.) when contrasted with his comprehension of concrete words (Experiment 6). In Experiment 7 a much wider range of categories was probed using the Battig and Montague (1969) stimulus lists. J.B.R.'s knowledge of this range of categories appeared to be distributed bimodally, both when compared with an 'expected' score and with the performance of a matched control subject; knowledge of most categories was either relatively spared or grossly impaired. The significance of these findings is discussed below. 2. Comparison of Verbal and Visual Modalities Although these patients were selected on the basis of visual identification diffi- culties they were found to have very comparable difficulties in comprehending the spoken word. Experiment 1, which provided a baseline measure of the severity of their visual identification difficulties for comparison with previously reported patients (Warrington, 1975,1981), also permitted a direct comparison of knowledge within the visual and verbal domains. Their performance was at a very similar level on both tasks. In the three experiments (2, 4 and 5) which explore category specificity in both domains, an identical pattern of selective preservation and impairment was observed. 3. Response Consistency Somewhat different patterns of responses were obtained for presentation of the same stimulus item on separate occasions within a modality and between modalities. In general, with one exception, there was an insignificant degree of consistency for comparison of responses between modalities (4/5). In contrast there was a signifi- cant degree of response consistency for all comparisons within a modality (3/3). 4. Semantic and Phonological Errors In Experiments 3 and 4 superordinate (categorical) information was preserved for a high proportion of the items on which subordinate information could not be CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 847 retrieved. In Experiment 6 that investigated knowledge of the full abstract/concrete range of content words there was also a high incidence of error responses which indicated preservation of superordinate information. In addition a small but significant number of the 'phonological' errors previously observed by Coughlan and Warrington (1981) were recorded.

Our discussion of these findings will concentrate on their relevance for understanding the organization of the semantic systems. But first, we must consider whether any of our findings could be attributed (1) to artefacts of stimulus selection or (2) to other cognitive deficits. Word frequency is a critical parameter for a wide range of verbal skills in the brain-damaged population (e.g. Howes, 1964; Poeck and Stachowiak, 1975). In our experiments, the test stimuli within each category were as far as possible balanced for word frequency. It is more difficult to balance for visual familiarity, but in Experiment 5 that compared food, animal, and inanimate object pictures, the effects of visual familiarity were partialled out by using the norms of Snodgrass and Vanderwart (1980); category effects were still obtained. Furthermore in a subsidiary experiment on normal subjects we found that the type of food stimuli used were rated the most familiar for a given word frequency (T. Shallice and R. McCarthy, unpublished). Thus one of the categories showing a robust and reliable deficit is also the most familiar. It is of course possible that past experience might also influence word frequency and visual familiarity for a particular individual. However, the present series consisted of patients of both sexes, with a wide age range (23 to 55 years), who had very varied educational and occupational histories. Furthermore they all suffered from the same viral illness that is known to affect structures bilaterally (Illis and Gostling, 1972) and this was the case for these patients. We would therefore reject any explanation of our central findings in terms of stimulus selection bias or individual variations of cerebral organization. Secondly, it is necessary to consider the possible confounding effects of associated deficits, in particular amnesia, dysphasia and perceptual impairments. There was evidence of a moderately severe amnesic syndrome in all 4 patients and it might be suggested that their retrograde deficit was so severe that information acquired in early life was not preserved. If, however, loss of knowledge acquired in childhood resulted from the same functional impairment as amnesia then any patient with a severe semantic memory deficit would also have to have a severe amnesic syndrome. This is not the case (see Warrington, 1975; Coughlan and Warrington, 1981). Furthermore, the category specificity findings themselves make the retrograde amnesia hypothesis untenable; any suggestion that food names were learned after names of inanimate objects would be absurd. Similarly, the category specificity findings are very difficult to explain in terms of some form of visual processing deficit. In view of the associated aphasic impairments recorded in all four of these 848 ELIZABETH K. WARRINGTON AND T. SHALLICE patients it is necessary to consider whether our results were confounded by word- finding difficulties. It is, however, unclear how the impoverished residual expressive vocabularies of our patients would bias their responses against living things and foods and in the case of S.B.Y. in favour of abstract words. In this regard it is worth noting that both J.B.R. and S.B.Y., in addition to being able to recognize colours, had an adequate colour name vocabulary. More critically on the picture/word matching tasks in which no verbal responses were required, equally clear evidence of category specificity was obtained for both the dissociation between the abstract and concrete categories (S.B.Y.) and that between inanimate objects, animals and foods (J.B.R., K.B., I.N.G.). In short, we claim that our findings are not secondary to an amnesic syndrome or to deficits of or expressive speech. It would appear therefore that the observed comprehension deficits arise within the semantic domain. Before considering in more detail the significance of category specific impair- ments we will discuss briefly whether the deficits should be attributed to a degradation of the semantic entries themselves or alternatively to difficulty in the process of accessing the full semantic representation of the concept. It has previously been argued by ourselves (Warrington and Shallice, 1979) that the hallmarks of a degradation deficit are consistency of responding across multiple presentations of the same stimulus, especially the vulnerability of subordinate information and marked frequency effects. These properties were recorded in the present investiga- tion. The relative invariance of identified and unidentified stimuli we interpret as reflecting a static impoverished knowledge base. Our data suggests that there is less correspondence between their knowledge of pictures and words—their visual and auditory vocabularies. Thus a particular concept can be consistently known from its visual representation and consistently not known from its verbal representation or of course the converse (the severity of the verbal and visual deficits being much the same). Strictly our arguments apply to the inanimate object category only, as floor and ceiling effects prevented a meaningful consistency analysis (both within and between modalities) for the other categories. Our inference is that two semantic systems—visual and verbal—are implicated and both are impaired in these patients. This conclusion is in accord with the findingsreviewe d in the Introduction that point to the existence of separate visual and verbal semantic systems. The major finding of interest in this study is the evidence for category specificity. We have argued that our findings provide compelling evidence for the specific impairment and preservation of semantic information. It then becomes necessary to specify and explain the observed dissociations between preservation of knowledge of inanimate objects and impairment of knowledge of living things and foods. In fact the converse dissociation has already been documented for the auditory modality in a patient (V.E.R.) in whom comprehension of inanimate object names was impaired, food, animal and plant names being preserved- (Warrington and McCarthy, 1983). If the categories sampled both in that investigation and the present study are considered, the overall pattern of performance appears to be the CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 849 same except that the preserved and the impaired categories are reversed. The existence of this double dissociation eliminates any explanation in terms of general task difficulty or the increased vulnerability of particular categories. In the case of V.E.R. it was argued that the preserved categories were those in which the identification of a particular category member depended upon a differentiation which stressed sensory features. To distinguish between a strawberry and a raspberry depends critically upon fine differences in colour, shape, size and texture. Similarly such information must also be accessed to achieve precise verbal comprehension. Inanimate objects unlike, say, most animals and plants, have clearly defined functions. The evolutionary development of tool using has led to finer and finer functional differentiations of artefacts for an increasing range of purposes. Individual inanimate objects have specific functions and are designed for activities appropriate to their function. Consider, for instance, chalk, crayon and pencil; they are all used for drawing and writing, but they have subtly different functions. Their crucial defining characteristic is their use with a particular writing surface, for example, blackboard, drawing paper, writing paper. Similarly, jar, jug and vase are identified in terms of their function, namely, to hold a particular type of object, but the sensory features of each can vary considerably. By contrast, functional attributes contribute minimally to the identification of living things (e.g. lion, tiger and leopard), whereas sensory attributes provide the definitive characteristics (e.g. plain, striped, or spotted). We would suggest that identification of an inanimate object crucially depends on determination of its functional significance, but that this is irrelevant for identification of living things. We would therefore speculate that a semantic system based on functional specifications might have evolved for the identification of inanimate objects. The other major categories we have considered might well have some representation in such a system, but this, we assume, would be inadequate for precise identification. To achieve precise identification for foods and living things, other semantic systems based on sensory features would be required (see Warrington and McCarthy, 1983). Such semantic systems would presumably have different patterns of associative links with other cognitive systems such as those underlying action and intention and associative links with other sensory modalities. This distinction between functional significance and sensory features is clearly applicable to the major categorical dissociation we have observed, namely, those between inanimate objects on the one hand and living things and foods on the other. How does this distinction relate to our findings on the wider range of categories sampled in Experiment 7? The observation of a bimodal distribution of perfor- mance between those categories that are relatively spared and those that are grossly impaired supports the concept of category specific preservation and impairment. Although the findings for many of the individual categories selectively preserved or impaired are in good accord with our major distinction, this formulation may well be incomplete. For instance the distinction is hardly applicable to categories such as weather and land formation (both preserved categories). We note that impaired 850 ELIZABETH K. WARRINGTON AND T. SHALLICE categories include cloths and precious stones, which are inanimate objects. We would argue, however, that these categories have only a 'generic' function and that such stimuli might be more easily differentiated by their sensory features. A more abstract category—diseases—can perhaps also be understood in these terms. The major anomalies to emerge are the findings that J.B.R. obtained a very poor score indeed on the musical instrument category and a relatively good score on body parts. Whether these results reflect the confounding effects of word frequency or whether there is a more fine grain categorical organization of semantic systems than has hitherto been supposed needs to be clarified by further studies. Nevertheless evidence for category specificity, per se, would appear to be robust. The dissocia- tions are striking and generalize over a wide range of stimuli. Further support for a more general level of category specificity within the semantic systems is provided by the findings on the comprehension of abstract words. The relative preservation of the comprehension of abstract concepts compared with concrete concepts has been previously reported (Warrington, 1975). S.B.Y. (Experiment 6) provides a very clear second instance of this unexpected syndrome. His comprehension of low frequency abstract words was nearly perfect, far superior to his comprehension of concrete words of equivalent frequency. (For further discussion of the significance of category specificity, see Warrington and McCarthy, 1983.) Neurological Considerations All 4 patients described here were diagnosed as cases of herpes simplex encephalitis. In each, there was evidence of bilateral brain disease affecting in particular the temporal lobe structures. The severe global amnesic sydrome present in all 4 patients has frequently been described in herpes simplex encephalitis and is well recognized to be associated with bilateral temporal lobe disease. The neuro- psychological syndrome we have described in this investigation—a specific impair- ment of visual and verbal identification—would generally be accepted as lateralized to the left hemisphere, the temporal and ventral occipital regions being considered critical structures (Hecaen and Albert, 1978; Coughlan and Warrington, 1978). The lesions in our patients were widespread but included these regions. Of particular neurological interest is that features of the Kliiver-Bucy syndrome are a not infrequent feature in the acute phase of herpes simplex encephalitis {see Hierons et al, 1978). Greenwood et al. (1983) have discussed this aspect of the illness in more detail and conclude that a more frequently observed symptom is bizarre eating and drinking behaviour. Three of our 4 patients were indeed observed to eat and drink indiscriminately in the acute phase. As we have documented a severe difficulty in identifying foods and comprehending food names in these patients the question arises as to whether their agnosia is causally related to the Kliiver-Bucy syndrome or is an associated deficit. We have described a new constellation of deficits in which, broadly speaking, CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 851 visual identification and verbal comprehension of living things and foods appear to be selectively impaired and inanimate objects selectively preserved. Further investigation will establish whether or not this is a commonly occurring syndrome in patients with herpes simplex encephalitis.

ACKNOWLEDGEMENTS

We wish to thank Dr L. J. Findley, Professor J. Marshall, the late Dr R. T. C. Pratt and Dr Michael Trimble for permission to investigate patients under their care and to report our findings. T.S. is a member of the Medical Research Council Applied Psychology Unit, and we wish to acknowledge the Council's support. We are grateful to J. Willison, C. Maciejewski and Dr K. Lewis who rated our subjects' responses. We particularly wish to thank Dr Ian Nimmo-Smith for his invaluable statistical advice and assistance.

REFERENCES

BATTIG W F, MONTAGUE W E (1969) Category norms for verbal items in 56 categories: a replication and extension of the Connecticut category norms. Journal of Experimental Psychology Mono- graph, 80, No. 3, Part 2. BEAUVOIS M-F (1982) Optic aphasia: a process of interaction between vision and language. Philosophical Transactions, Royal Society, London, B, 298, 35-47. BEAUVOIS M-F, SAILLANT B, MHNINGER V, LHERMITTE F (1978) Bilateral tactile aphasia: a tacto-verbal dysfunction. Brain, 101, 381-401. BENSON D F (1979) Aphasia, Alexia and Agraphia. New York: Churchill Livingstone. CARROLL J B, DAVTES P, RICHMAN B (1971) Word Frequency Book. Boston, Massachusetts: Houghton Mifflin. COUGHLAN A K, WARRINGTON E K (1978) Word-comprehension and word-retrieval in patients with localized cerebral lesions. Brain, 101, 163-185. COUGHLAN A K, WARRINGTON E K (1981) The impairment of verbal semantic memory: a single case study. Journal of Neurology, Neurosurgery and Psychiatry, 44, 1079-1083. EvERiTT B S (1977) The Analysis of Contingency Tables. London: Chapman and Hall. GESCHWIND N (1965) Disconnexion syndromes in animals and man. Part II. Brain, 88, 585-644. GOODGLASS H, KLEIN B, CAREY P, JONES K (1966) Specific semantic word categories in aphasia. Cortex, 2, 74-89. GREENWOOD R, BHALLA A, GORDON A, ROBERTS J (1983) Behaviour disturbances during recovery from herpes simplex encephalitis. Journal of Neurology, Neurosurgery and Psychiatry, 46, 809-817. HECAEN H, DE AJURIAGUERRA J (1956) Agnosie visuelle pour les objets inanimes par lesion unilateral gauche. Revue Newologique, 94, 222-233. HECAEN H, ALBERT M L (1978) Neuropsychology. New York: John Wiley. HECAEN H, GOLDBLUM M C, MASURE M C, RAMIER A M (1974) Une nouvelle observation d'agnosie d'objet. Deficit de l'association, ou de la categorisation, specifique de la modalite visuelle? Neuropsychologia, 12, 447-464. HIERONS R, JANOTA I, CORSELLIS JAN (1978) The late effects of necrotizing encephalitis of the temporal lobes and limbic areas: a clinico-pathological study of 10 cases. Psychological Medicine, 8,21-42. HOWES D (1964) Application of the word frequency concept to aphasia. In: Ciba Foundation Symposium on Disorders of Language. Edited by A. V. S. de Reuck and M. O'Connor. London: Churchill. 852 ELIZABETH K. WARRINGTON AND T. SHALLICE

Tins L S, GOSTUNG J V T (1972) Herpes Simplex Encephalitis. Bristol: Scientechnica. KONORSKJ J (1967) Integrative Activity of the Brain: an Interdisciplinary Approach. Chicago and London: University of Chicago Press, pp. 115-120. MCKENNA P, WARRTNGTON E K (1983) Graded Naming Test. Windsor, England: NFER-Nelson. NELSON H E (1982) The National Adult Reading Test. Windsor, England: NFER-Nelson. NIELSEN J M (1946) Agnosia, Apraxia, Aphasia. Their Value in Cerebral Localization. New York: Hoeber. POECK K, STACHOWIAK F-J (1975) Farbbenennungsstorungen bei aphasischen und nichtaphasischen Hirnkranken. Journal of Neurology, 209, 95-102. RATCLIFF G, NEWCOMBE F (1982) Object recognition: some deductions from clinical evidence. In: Normality and Pathology in Cognitive Functions. Edited by A. W. Ellis. London: Academic Press. RUBENS A B, BENSON D F (1971) Associative visual agnosia. Archives of Neurology, Chicago, 24, 305-316. SIEGEL S (1956) Nonparametric Statistics. New York: McGraw-Hill. SiLVERMAN B W (1981) Using kernel density estimates to investigate multimodality. Journal of Royal Statistical Society, B, 43, 97-99. SNODGRASS J G, VANDERWART, M. (1980) A standardized set of 260 pictures: norms for name agreement, image agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human Learning and Memory, 6, 174-215. WARRINGTON E K (1975) The selective impairment of semantic memory. Quarterly Journal of Experimental Psychology, 27, 635-657. WARRINGTON E K (1981)Concrete word dyslexia. British Journal of Psychology, 72, 175-196. WARRINGTON E K (1984) Recognition Memory Test. Windsor, England: NFER-Nelson. WARRINGTON E K, JAMES M (1967) Disorders of visual perception in patients with localized cerebral lesions. Neuropsychologia, 5, 253-266. WARRINGTON E K, MCCARTHY R (1983) Category specific access dysphasia. Brain, 106, 859-878. WARRINGTON E K, SHALLICE T (1979) Semantic access dyslexia. Brain, 102,43-63. WARRINGTON E K., TAYLOR A M (1978) Two categorical stages of object recognition. Perception, 7, 695-705. ZINKIN S, BIRTCHNELL J (1968) Unilateral electroconvulsive therapy: its effects on memory and its therapeutic efficacy. British Journal of Psychiatry, 114, 973-988. (Received August 18, 1983. Revised January 3, 1984)

(See overleaf for Appendix) CATEGORY SPECIFIC SEMANTIC IMPAIRMENTS 853

APPENDIX Stimuli used m Experiment 2 Animate ittrrndl

Freq. Freq Freq Peacock 33 Hedgehog 11 Pud* 9 Deer 316 Lobxter 29 Palm 119 Rose 461 Daffodil 6 Buttercup 4 Chestnut 24 Frog 171 Crab 62 Duck 216 Ladybird 2 Acorn 10 Camel 72 Holly 24 Wasp 32 Giraffe 40 Owl 137 Shixp 464 Kangaroo 43 Bar 557 Rabbit 209 Crocodile 61 Tuhp 20 Ostrich 33 Plrrot 94 Mouse 207 Beetle 41 Butterfly 77 Foxglove 0 Rhino 14 Tortoise 17 Ed 24 Shark 92 Foi 163 Poppy 15 Seaweed 41 Donkey 156 Toadstool 4 Squirrel 88 Crocus 0 Snail 47 Swan 29 Flamingo 7 Lizard 54 Spider 229

Inanimate stimuli

Freq Freq. Freq. Car 1752 Lighthouse 51 Scales 185 Submarine 122 Tent 208 Spade 7 Spanner 0 Decanter 0 Mittens 3 Thermometer 154 Kettle Bnefcue 3 Torch 28 Stool 58 Tankard 2 Violm 106 Towel 48 Wallet 13 Vice 14 Mincer 0 Axe 27 Piano 418 Toaster 16 Typewriter 43 Ladder 127 Scissors 72 Calculator 4 Saw 2900 Barometer 42 Binoculars 20 Drill 136 Canoe 164 Hummer 155 Tenni* racket 24 Trumpet 77 Compass 210 Wheelbarrow 21 Windmill 33 Tricycle 10 Dustbin 0 Slippers 38 Headphones 3 Hoover 22 Guitar 101 Pram 0 Mop 16 Umbrella 51 Drums 128

The word frequenaa were Uken from the Carroll et al (1971) count Where appropriate the frequency of the Amencan equivalent M given

Stimuli used in Experiment 4

Food Freq. Inanimate Freq. Grapefruit 18 RoUenkates 2 Mixed nuts 143 Paintbox 3 Cabbage 139 Swing 89 Pineapple 39 Balloon 233 Cake 244 Mirror 209 Cucumber 6 Camera 166 Beer 24 Crayon 83 Bread 515 Bicycle 182 Eggs 785 Cotton reel 12 Chocolate 143 Umbrella 51 Meat 617 Wheelbarrow 21 Tomatoes 49 Clothes peg 4 Orange 221 Pram 3 BiscmU 45 Button 130 Cheese 236 Mitten 3 Pear 32 Sandal 5 Icecream 33 Bdl 174 Soup 184 Double bass 2 Milk 849 Skipping rope 1 Sprouts 8 Trumpet 77 Lemon 44 Tankard 2 Strawberry 26 Basin (sink) 149 Baked beans 199 Radiator 25 LoDy 8 Shears 30 Onion 42 Electric Sre _ Mushroom 43 Scales 185 Chips 42 Jigsaw 8 Peas 70 Shot 182 JeDy 87 Drums 128 Apples 284 Tie 234